Inference on Taxonomy Subset

[mikepianka]

Thanks for this great repo! I have a question about running the universal demo on a subset of the universal taxonomy classes.

Is there a simple way to run the demo and inference task but only have it identify and segment a subset of classes from the universal taxonomy?

For example, only have it try identifying people in the input image and nothing else. I am curious if this would 1) speed up the run time since it is only trying to identify one class and 2) possibly improve the results if there are overlapping classes like a person behind a picket fence. With the default settings the fence tends to get identified in the segmentation and not the person behind it.

Thanks!